Beginner
What is the EU AI Act?
The EU AI Act is a European Union regulation establishing a risk-based framework for AI systems and, in some cases, general-purpose AI models. It defines categories of AI use and assigns obligations related to safety, transparency, governance, and compliance according to the level of risk.
Who is a provider under the EU AI Act?
A provider is a person, organization, public authority, agency, or other body that develops an AI system or general-purpose AI model, or has one developed, and places it on the market or puts it into service under its own name or trademark.
How is a provider different from a deployer?
Use Provider for the actor that develops, places on the market, or puts into service an AI system or GPAI model under its name; use Deployer for the actor that uses an AI system under its authority.
What is conformity assessment for an AI system?
Conformity assessment is the process of determining whether a system, product, or process satisfies specified legal or technical requirements.
What is LLM security?
LLM, or large language model, is a language model with a large number of parameters trained on broad text or multimodal data to understand and generate language.
What is prompt injection?
Prompt injection is an adversarial technique that manipulates a language model by placing malicious or unintended instructions in user input, documents, web pages, or other content the model processes.
What is a jailbreak in an LLM system?
Prompt injection or jailbreak refers to adversarial input intended to bypass a model's instructions, safety policies, or operational controls. Mitigations include input filtering, retrieval isolation, least-privilege tool access, policy enforcement, and layered guardrails.
How is RAG different from fine-tuning?
Use RAG when the model needs current, source-grounded information; use fine-tuning when the model needs specialized behavior, task performance, domain adaptation, or style learned from data.
What is AI governance?
AI Governance is the framework of policies, processes, roles, controls, and oversight used to steer AI development and use across its lifecycle. It aims to ensure responsible, lawful, secure, transparent, and accountable AI practice within an organization or jurisdiction.
What is an AI management system?
An AI Management System is an organizational platform or structured set of processes for managing AI projects and systems throughout their lifecycle. It coordinates governance, data flows, model development, deployment, monitoring, compliance, and accountability activities.
What is ISO/IEC 42001?
ISO/IEC 42001 is an international standard for establishing, implementing, maintaining, and improving an AI management system. It helps organizations govern AI development and use through policies, roles, risk management, oversight, and continual improvement.
What is an AI inventory system register?
An AI Inventory or System Register is an organized catalogue of AI systems and use cases within an organization. It typically records owners, purposes, data sources, deployment status, risks, controls, and other information needed for governance, audit, and accountability.
What is personal data in AI?
Personal data is information relating to an identified or identifiable natural person under data protection law. In EU contexts, its meaning is tied to Article 4(1) of Regulation (EU) 2016/679 and related legal interpretation.
What is special-category data in AI?
Special categories of personal data are highly sensitive types of personal data identified in EU data protection law.
Is biometric data personal data?
Biometric data is personal data produced by specific technical processing of a person's physical, physiological, or behavioral characteristics. Examples include facial images, fingerprints, voice patterns, gait, and other identifiers that can support recognition or identification.
What is anonymisation in AI data governance?
Anonymisation and de-identification are techniques used to reduce the risk that individuals can be identified from a dataset. In practice, irreversible anonymisation can be difficult, especially when datasets can be linked with other information.
What is AI safety?
AI Safety is the interdisciplinary field and set of practices focused on preventing harm from AI systems. It includes reliability, misuse prevention, bias and error mitigation, alignment with human values, and analysis of both near-term and long-term risks.
How is risk different from harm?
Use risk for potential adverse outcomes before or during assessment; use harm for the adverse effect that has occurred or may be concretely described.
What is a safety case for AI?
An AI safety case is a documented argument, supported by evidence, that an AI system is acceptably safe for a defined use and operating context. It is especially relevant for higher-risk sectors where assurance must be explicit, reviewable, and maintained over time.
What is model drift?
Model drift is the degradation or change in model performance over time as real-world data, user behavior, or operating conditions diverge from the data used during training or evaluation. Detecting drift is important for deciding when to retrain, recalibrate, or replace a model.
What is model evaluation?
Evaluation is the process of measuring the quality, behavior, or performance of a model, system, or change against defined criteria. In machine learning it may use validation and test data, while LLM evaluation may also include safety, factuality, robustness, and user-impact assessments.
What is a benchmark in AI evaluation?
A benchmark is a standardized test, dataset, task, or evaluation procedure used to measure and compare the performance of AI systems.
How are precision and recall different?
Use Precision when false positives are costly; use Recall when missed positives are costly.
Why can accuracy be misleading?
Accuracy is a classification metric equal to the proportion of predictions that are correct among all predictions made. It is useful in some evaluation settings but can be misleading when classes are imbalanced, where precision, recall, or other metrics may better reflect model performance.
What is agentic AI?
Agentic AI refers to AI systems that can pursue objectives by perceiving context, making decisions, planning steps, and taking actions with some degree of autonomy.
What is an AI agent?
An AI Agent is an entity or software system that perceives an environment, interprets inputs, and takes actions to achieve goals.
What is an agentic workflow?
An Agentic Workflow is a process in which an AI agent plans and carries out multiple steps toward a defined goal. It may include reasoning, tool invocation, external system interaction, monitoring, and self-correction as conditions change.
What is an agentic loop?
An Agentic Loop is the repeated cycle through which an AI agent observes a situation, reasons about what to do, acts, and uses feedback to continue or stop. The loop continues until a goal is achieved, a failure condition occurs, or another termination rule is met.
What is biometric AI?
Biometric technology refers to systems that use biometric data to recognize, verify, or identify a person's identity. Examples include facial recognition, fingerprint scanning, iris recognition, voice recognition, and related identity-assurance tools.
What is biometric data?
Biometric data is personal data produced by specific technical processing of a person's physical, physiological, or behavioral characteristics. Examples include facial images, fingerprints, voice patterns, gait, and other identifiers that can support recognition or identification.
What is biometric identification?
Biometric identification is the automated recognition of a natural person's identity by comparing that person's biometric data against biometric data stored in a database. It is typically a one-to-many matching process used to establish who the person is.
How is biometric verification different from biometric identification?
Use Biometric Identification when the focus is automated recognition of a natural person's identity by comparing that person's biometric data against biometric data stored in a database; use Biometric Verification when the focus is automated one-to-one confirmation of a person's identity by comparing their biometric.
What is machine learning?
Machine Learning (ML) is a subfield of artificial intelligence in which systems learn patterns from data or experience to make predictions, classifications, recommendations, or decisions without being explicitly programmed for every case.
What is supervised learning?
Supervised learning is a machine learning approach in which a model learns from examples that pair inputs with known target outputs or labels. The model uses those labeled examples to learn mappings that can support prediction, classification, or regression on new data.
How is unsupervised learning different from supervised learning?
Use Supervised Learning when labeled input-output examples exist; use Unsupervised Learning when the goal is to discover structure without labels.
What is reinforcement learning?
Reinforcement Learning (RL) is a machine learning paradigm in which an agent learns to act in an environment by receiving rewards or penalties for its actions. The objective is to learn a policy that maximizes expected cumulative reward over time.
What is generative AI?
Generative AI (GenAI) refers to AI systems that create new content, such as text, images, audio, video, code, or other media, in response to prompts or context. These systems learn patterns from training data and generate outputs that resemble or transform the kinds of content they were trained on.
How is generative AI different from artificial intelligence?
Use Artificial Intelligence for the broad field and system category; use Generative AI when the system creates new content from prompts or context.
What is a large language model?
LLM, or large language model, is a language model with a large number of parameters trained on broad text or multimodal data to understand and generate language.
What is a diffusion model?
A diffusion model is a generative model trained to reverse a process that gradually adds noise to data. By learning to denoise from random noise toward a target data distribution, it can generate realistic images, audio, text representations, or other outputs.